A study on Quadruped Firefighting Robot Tactics for Supporting Life Search and Fire Suppression Activities at Firefighting Sites
Article information
Abstract
This study was conducted as part of a project by the Ministry of Trade, Industry, and Energy to develop sensor and robot technologies for firefighting. It aims to evaluate and enhance the operational capabilities of the Unitree Go2 Pro Quadruped robot in various firefighting scenarios. Despite its limitations in disaster response, the robot was tested to derive effective firefighting strategies. The findings revealed that the robot's performance on stairs and slopes was insufficient for field deployment; however, critical requirements for firefighting robots were identified. The inclusion of follower mode and obstacle-jumping capabilities emerged as essential for developing comprehensive tactics. Impact resistance was successfully verified through physical tests, whereas heat and fire resistance could only be assessed through simulations. Communication analysis underscored the necessity of voice broadcasting and highlighted limitations in detecting glass walls and smoke. Furthermore, significant signal degradation through walls emphasized the need for improvements in future developments.
1. Introduction
Annual fire incidents result in casualties and increased property damage. From 2014 to 2023, injuries and fatalities consistently reached 2000 each year, highlighting the growing impact of fires[1]. Although major fires represented only 0.028% of all fires during this period, they inflicted disproportionate harm, accounting for 8.07% of casualties and 21.5% of property damage[1]. In 2021 alone, major fires accounted for more than half of all fire-related property damage, underscoring their severe financial impacts.
The rise in injuries among on-duty firefighters underscores the harsh realities of fire scenarios. Despite wearing heat-resistant suits and other protective gear, firefighters face a high risk of developing musculoskeletal disorders due to the physically demanding nature of their roles[2]. These roles include structural demolition, emergency patient transport, and the handling of heavy equipment such as fire hoses and firefighting gear. This study aims to introduce and evaluate a model that employs firefighting robots in fire scenes for human searches and fire suppression.
The rapid advancement of South Korea's logistics industry has led to a substantial increase in the number of large-scale warehouse facilities. Over the past three years (2019–2023), the number of medium-sized warehouses (approximately 16,500-33,000 square meters) has increased by approximately 96%, whereas large warehouses (33,000-99,000 square meters) have grown by approximately 165%[3]. Furthermore, warehouses exceeding 99,000 square meters have doubled during the same period[3]. In the event of a large warehouse fire, the considerable fire load and the vertical stacking of combustible materials pose major challenges for fire suppression. The accumulation of dense black smoke within the facility further exacerbates visibility issues, making navigation extremely difficult and increasing the likelihood of disorientation[4]. Notably, a firefighter fatality incident in South Korea occurred despite the use of a lifeline for orientation; firefighters were unable to locate the exit due to severe smoke conditions, which resulted in casualties.
Given the current trend of population decline in South Korea and the increasing challenges of securing sufficient human resources, there is a growing need to explore the use of quadruped robots as substitutes for human firefighters in hazardous environments.
To address this need, we analyzed and tested the physical functions, heat resistance, durability, and communication capabilities necessary for deployment in fire scenes and for the development of firefighting robot tactics. We examined physical features such as stair climbing, battery efficiency, operational range, follow mode, and auto-recovery. Durability functions were derived through experiments conducted with an actual quadruped robot. Furthermore, communication capabilities were investigated through experiments assessing the robot's voice function, command reception, essential strategic and tactical functions, and ability to overcome communication distance limitations. This Study aims to develop a quadruped robot for firefighting by testing the fundamental performance required for its deployment in fire scenarios.
2. Firefighting Robot Tactics
To assist firefighters in urgent fire situations, the physical properties, heat resistance, durability, and communication capabilities of a quadruped robot were experimentally analyzed. The capabilities of the robot were examined using a commercially available Unitree Go2 Pro model (Figure 1). Based on these evaluations, research was conducted to establish the requirements for implementing the robot in firefighting scenarios.
The unitree Go2 pro quadruped robot (70 cm × 31 cm × 40 cm; ~ 15 kg) used in the experiments.
2.1. Physical Functions
To experimentally investigate the physical functionalities, the detailed features required for the deployment of the quadrupedal robot in fire scenes were systematically organized and analyzed, as illustrated in Figure 2. Accessibility tests were conducted on 18-cm high stairs and 15-degree slopes, which are commonly specified in commercial buildings, to verify the robot’s ability to enter buildings through various access points, as depicted in Figure 3.
Mind map for deriving specifications of physical functions.
(a) Stairways and (b) slopes at a business facility in Seongdong-gu.
Battery endurance was evaluated by analyzing operational duration across different terrains—flat surfaces, stairs, and slopes—under typical ambient temperature conditions (25 ℃). This analysis was complemented by assessing endurance relative to variations in speed, as shown in Figure 4. Additional load tests with weights of 5 and 10 kg were conducted to determine the impact of varying load conditions on battery life.
Walking speed of the quadruped robot in manual mode.
Obstacle detection capabilities were examined to assess the robot’s ability to autonomously navigate around obstacles such as columns and people within buildings, which is essential for effective maneuvering in fire scenes. The robot's capacity to maintain a consistent distance from firefighters during operation was tested to evaluate its performance in follower mode. This included assessing the turning radius based on robot-human spacing during navigation through doors and stairs, conducted at the same location as the stair navigation experiments.
The recovery functionalities embedded in the firefighting quadruped robot were analyzed to establish the necessary operational requirements. Finally, the force required to open doors with the robotic arm was measured for both standard and fire doors, as shown in Figures 5(a) and 5(b), using the force measurement device depicted in Figure 5(c).
(a) Standard door, (b) fire door, and (c) closing force measurement device at a business facility in Seongdong-gu.
2.2 Durability Functions
The Unitree Go2 Pro, a test model designed for general environments rather than industrial hazard applications, is constructed from aluminum and plastic. This presented challenges in conducting fire resistance, heat resistance, waterproofing, and dustproofing tests. Consequently, the durability specifications were derived solely from impact resistance tests.
Durability tests were conducted using limited experimental methods. Light collisions, simulating a firefighter bumping into the robot while moving, were assumed, and impact resistance tests were performed on all four sides. As no established standards exist for external collisions with quadruped robots, a pushing force of approximately 400 N, defined as the one-time pushing force limit in ISO 11228-2, was applied during the tests. The robot maintained stability and operated normally without falling. These experiments were also conducted with added loads identical to those used in the physical function tests (5 and 10 kg), taking into account the robot's maximum payload of 10 kg.
2.3 Communication Functions
Communication capabilities are essential for robot operations. The specifications required for deployment in fire scenes were derived, including physical communication distance and methods for overcoming communication range limitations in smoke-filled environments, as illustrated in Figure 6.
Mind map for deriving specifications of communication functions.
To evaluate the robot’s ability to transmit and receive messages crucial for coordinating rescue operations and delivering firefighter commands, decibel levels were measured at distances of 1 m, 5 m, and 10 m from the robot using a Decibel Meter. These measurements were intended to confirm the survival status of individuals requesting rescue and to facilitate communication between firefighters and robots under various conditions.
In addition, to assess the feasibility of voice transmission through the robot in noisy environments, tests were conducted in an area with a background noise level of 100 dB. This experiment was designed to evaluate the effectiveness of the robot in conveying commands and information in high-noise scenarios typically encountered during firefighting operations.
Experiments were conducted using a lighting system installed on the robot to facilitate the identification of individuals requesting rescue via video transmission. The robot’s light-emitting device exhibited a maximum brightness of 9700 LUX. Beginning with the lights turned off, the brightness was incrementally increased by 10% at each step to determine the levels necessary for effective human search operations. This procedure was designed to optimize the robot’s ability to identify individuals under varying lighting conditions, thereby enhancing its effectiveness in search and rescue missions.
Tests were conducted to evaluate the performance degradation of the robot as battery capacity decreased. At 20% and 10% remaining battery life, the impact on various functions—including speed, voice output, battery consumption rate, light brightness, and communication speed—was measured. The results were used to assess how each capability was affected by reduced battery charge, providing insights into the operational limits under low-battery conditions.
Tests were conducted to determine the operable transmission range required for maneuvering the robot. The distance between the robot and the receiver was varied from 5 to 150 m to measure data transmission speed, based on data from the quadruped robot control application. From these tests, the maximum operational distance at which normal operations were possible, the distance at which communication became unreliable, and the distance at which the quality of video and sensor outputs degraded were derived. Additionally, the impact of physical barriers, such as walls (Figure 7), on communication quality between the operator and the robot was assessed. The maximum control distances for both router-based and Wi-Fi communications were also determined.
Fire door at a business facility in Seongdong-gu.
In the final stage of the communication function tests, sensor accuracy was evaluated by assessing the robot’s ability to detect the presence or absence of glass walls. In addition, a smoke generator was used to produce smoke, and the extent to which the sensors could detect it was assessed.
3. Results
3.1 Experimental Results of Physical Functions
3.1.1 Stair Climbing and Descending
The tested robot demonstrated a step-climbing capability of 12 cm, which was insufficient for smoothly ascending stairs. When equipped with firefighting equipment, a robotic arm, and a water hose, the additional load further hindered mobility, highlighting the need for improvements in firefighting robots. The speed of ascending stairs stairs was 0.1 m/s, whereas a performance of 0.4 m/s is required to effectively assist firefighters in Table 1. as referenced in[5,6]. Descending stairs was not possible, and manual adjustments to the walking algorithm were necessary even for ascending. These limitations suggest that significant improvements are necessary for the robot’s actual deployment in fire-site scenarios.
Speed Comparison During Stair Climbing
3.1.2 Slope Maneuvering
The robot demonstrated a significant reduction in speed when traveling uphill on a slope, while maintaining stable operation. Similarly, downhill travel remained stable; however, instability was observed when carrying a load. This instability may cause delays during deployment from the TCMV at the scene, indicating the need for improvement.
3.1.3 Battery Duration
The quadruped robot was operated at a constant speed to evaluate the efficiency of its 8000 mAh battery. The classification of robot speeds is presented in Figure 4.
The experimental results in Table 2. indicated that a speed of 1.25 m/s was the most efficient. Given that the average human walking speed is 1.33 m/s, a speed of 1.25 m/s is appropriate for a robot to follow firefighters in the field, balancing operational capability and battery efficiency.
Battery Depletion Time According to Speed
Battery efficiency experiments were also conducted under various load conditions, and additional experiments were conducted to measure battery consumption during stair climbing.
Battery efficiency experiments were conducted under various load conditionsIn a, along with tests to measure the battery consumption time during stair climbing. The results showed in Table 3. a 35.4% reduction in battery depletion time when a 10 kg load was applied compared to normal conditions. During stair ascent, battery depletion time decreased by 39.9% compared to operation on flat terrain. Battery efficiency experiments were conducted under various load conditions, along with tests to measure battery consumption time during stair climbing. The results showed a 35.4% reduction in battery depletion time when a 10 kg load was applied compared to normal conditions. During stair ascent, battery depletion time decreased by 39.9% compared to operation on flat terrain.
Required Time According to Load Capacity and Ascending Stairs
3.1.4 Obstacle Detection/Overcoming and Automatic Recovery
Experiments were conducted to evaluate the robot’s ability to detect obstacles using a LiDAR sensor. As shown in Figure 8, the robot successfully recognized columns and people as well as distinguished between open and closed doors.
LiDAR sensor recognition of a (a) column, (b) people, (c) closed door, and (d) open door.
Additional Experiments were performed to assess the obstacle-overcoming capabilities of the quadruped robot using its jumping function. The results showed a maximum jump height of 30 cm and a maximum jump distance of 80 cm. Given the step-climbing limit of 18 cm, overcoming obstacles by jumping was deemed feasible.
Experiments on the automatic recovery function showed that although the robot was capable of recovering from falls, it could not independently recognize when a fall had occurred. Additionally, while the recovery function performed effectively on flat terrain, it failed to restore the robot’s posture on uneven surfaces.
3.1.5 Follower Mode
In the follower mode, activating the avoidance function caused the robot to fail in passing through a 90 cm wide door, as the door was recognized as an obstacle. Experiments were conducted to determine the appropriate distance between the operator and the robot. As shown in Figure 9, configurations (a) and (b) failed to pass through the door, whereas configurations (c) and (d) experienced interference between the operator and the robot. Configurations (e) and (f) showed no interference with the robot, door, and operator, and encountered no issues when navigating stairs and corridors. A rear-following distance of 50 cm was determined to be the most suitable.
(a, b) Beside-follow at 50 cm, (c, d) beside-follow at 10 cm, and (e, f) rear-follow at 50 cm.
3.1.6 Robotic Arm
To derive the specifications of the robotic arm, the forces required to open both fire doors and regular doors were measured using a closing force measurement device in Table 4.
Force Required to Open the Door
During the operation of the smoke ventilation equipment, it was determined that the firefighting quadruped robot must exert opening and closing forces of at least 110 N to open the doors.
3.2 Experimental Results of Durability Functions
3.2.1 Physical Impact Test
The robot demonstrated resistance to falling when subjected to physical impacts on all four sides. However, the experimental results showed a noticeable decrease in stability when the robot was carrying a load. Additionally, when transporting liquids, fluid movement within the container caused continuous shifts in the robot's center of gravity, resulting in instability. Due to the characteristics of the walking algorithm, the robot also exhibited instability during movement under these conditions.
3.3 Experimental Results of Communication Functions
3.3.1 Voice Broadcasting Feature
The remote voice broadcasting feature was tested using a speaker mounted on athe quadrupedal robot. The results for voice broadcasting capability of a quadruped robot by distance in Table 5. Decibel measurements at various distacne revealed a reduction rate of 1 dB/m. Additionally, tests conducted in a noisy environment with 100 dB background noise confirmed that the operator's voice commands could still be communicated effectively.
Voice Broadcasting Capability of a Quadruped Robot by Distance
3.3.2 Strategic and Tactical Functions Required
The onboard lighting of the robot was also tested at various brightness levels. As shown in Figure 10, at 10% brightness (970 LUX), the visibility range was approximately 5 m, making object identification difficult. At 50% brightness (approximately 5000 LUX), objects were easily identifiable.
Visibility by brightness: (a, b) power off, (c, d) 10% power, and (e, f) 50% power.
Tests were conducted to evaluate the performance degradation of the robot at low battery levels. The results showed no reduction in speed, voice output, lighting brightness, or communication speed. However, battery consumption increased as the battery level decreased. Additionally, when communication was lost, the robot remained stationary until the connection was reestablished.
3.3.3 Operation Range Limitations
Results from the indoor communication experiment showed no degradation in the quality of the LiDAR sensors and videos up to a distance of 100 m. Although variations in data transmission speed were observed, they did not significantly affect the quality of the video and LiDAR sensors. The maximum communication distance was 150 m, beyond which the video experienced a rapid decline in frame rate, making control impossible, as shown in Figure 11.
(a) Distance between robot and operator outdoors, and (b) the reconnecting feature.
The results for data transmission speed as a function of distance are summarized in Table 6. These measurements were obtained using a router connection. When connected via Wi-Fi, the communication distance iwas approximately 40 m, consistent with the typical Wi-Fi range; however, this may vary depending on the Wi-Fi performance
Data Transmission Speed by Distance
Additionally, when a single wall was present between the operator and the robot, the maximum operational distance was reduced to 18.5 m, representing an 87.7% decrease from the original maximum distance. Communication was not possible when two walls were present.
3.3.4 Sensor Accuracy
Although glass walls were visually identified by the operator through the transmitted video, they were not detected by the LiDAR sensor (Figure 12). In the smoke detection experiments, smoke was visible in the video but was not recognized by the LiDAR sensor.
(a) Room filled with smoke, (b) robot in front of glass wall, and (c) LiDAR sensor display.
These limitations highlight the need for multimodal sensing systems, such as thermal imaging cameras and smoke detection sensors, to effectively recognize non-reflective or low-reflectivity obstacles, including glass, smoke, and flames. In particular, the inability to detect glass in building fire environments poses a critical challenge for real-world applications, as it may lead to navigation failures or collisions under smoke-obscured conditions.
4. Conclusions
As part of the Ministry of Trade, Industry, and Energy’s project to develop sensor and robot technology for firefighting, this study employed Unitree's Go2 Pro to inform firefighting tactics. Despite its limitations in disaster response, the robot proved suitable for testing under controlled conditions.
This study identified the key requirements for field-deployable firefighting quadruped robots, while demonstrating that the current performance on stairs and slopes remains insufficient for immediate deployment. Follower mode was found to be an essential capability, and the robot’s obstacle-clearing (jumping) behavior contributed to the development of comprehensive operational tactics. Mechanical robustness was partially verified; impact resistance was demonstrated through physical tests, whereas heat and flame resistance were assessed only through simulation owing to safety and facility constraints. Communication analysis highlighted the operational importance of voice broadcasting and revealed substantial signal attenuation through walls. The sensing performance further revealed limitations in detecting glass walls and smoke with LiDAR/RGB sensors, underscoring the need for multimodal sensing, such as thermal imaging and smoke detection sensors, to reliably recognize glass, smoke, and flames. Based on these findings, system requirements were specified for quadruped platforms that integrate human detection and fire suppression modules, which were then translated into comprehensive robot tactics for fire scenes. To enable deployment in actual fire scenarios, research on fire and heat resistance must be conducted in advance. Moreover, because the experiments were not conducted in actual fires or high-fidelity simulated environments, the generalizability of the results is limited. Rigorous validation in controlled mock settings and post-suppression field trials is therefore required.
Notes
Author Contributions
Conceptualization, Kyeongmin Kim and Eunsoo Son; Methodology, Kyeongmin Kim; Software, Kyeongmin Kim; Validation, Kyeongmin Kim and Eunsoo Son; Formal analysis, Kyeongmin Kim; Investigation, Kyeongmin Kim; Resources, Eunsoo Son; Data curation, Kyeongmin Kim; Writing—original draft preparation, Kyeongmin Kim; Writing—review and editing, Kyeongmin Kim; Visualization, Kyeongmin Kim; Supervision, Eunsoo Son; Project administration, Eunsoo Son; Funding acquisition, Eunsoo Son. All authors have read and agreed to the published version of the manuscript.”
Conflicts of Interest
Kyeongmin Kim and Eunsoo Son are employed by KF UBIS. Co., Ltd.; however, this relationship did not influence the study’s results or conclusions.
Acknowledgments
This research was supported by the Ministry of Trade, Industry and Energy / Korea Evaluation Institute of Industrial Technology (KEIT) as part of the “Development of Safety Robot Technology for Accident Prevention of Workers in Disaster and Hazardous Work Sites” project (Grant no. 20026194). We appreciate the support and funding of this study.